Free tool
LUFS loudness meter
Measure integrated loudness, loudness range, crest factor and peak level for any audio file, then see what each streaming platform would do to it on playback. Decoding and measurement happen in your browser — the file never leaves your machine.
Short answer
Can a loudness measurement tell you a track is AI-generated?
No. Generated tracks usually arrive loudness-maximised, but so does most commercial music — the numbers describe mastering decisions, not origin. Loudness is useful here for the opposite reason: knowing a master is heavily limited tells you why a detector score may read high on a perfectly human recording.
Where detection misleads →Drop an audio file here
MP3, WAV, FLAC, M4A, Ogg or Opus. Files longer than five minutes are measured on the first five.
What the four numbers mean
Integrated loudness is one figure for the whole track in LUFS, measured the way broadcasters and streaming services measure it: the signal is K-weighted to approximate how ears respond, chopped into 400 ms blocks, and quiet blocks are gated out so silence and fades do not drag the average down. It is the number platforms compare against their target.
Loudness range (LRA) describes how much the level moves between sections, in loudness units. A ballad with a whispered verse and a full chorus has a wide range; a club track engineered to sit at one level has a narrow one. Crest factor is the distance between the loudest peak and the average level, which is the most direct measure of how hard a limiter has been working. Sample peak is simply the highest sample value, in dBFS.
Why a loudness meter sits on a detection site
Because dynamics are the most misused evidence in this whole field. The claim that AI music "sounds flat" or "has no dynamics" is repeated constantly, and it collapses the moment you measure real material: commercial masters in pop, hip-hop and electronic music routinely show crest factors under 8 dB, which is exactly where generated tracks land. That overlap is not a coincidence — generators were trained on those masters and most apply their own limiting before delivery.
So the useful move is to measure the mastering first, then read the detection score with that context in hand. If this tool tells you a file is limited to within 6 dB of its average level, a high probability from an acoustic detector deserves less weight, because one of the behaviours it relies on has been overwritten by a mastering chain. We set out the same argument with numbers in the electronic-music study and in what your result means.
Practical uses that have nothing to do with AI
Most people arrive here for delivery questions, and the tool answers those directly. Check where a master lands before uploading, compare a mix against a reference track, confirm that you left headroom before lossy encoding, or verify that a mastering engineer's revision actually changed level rather than just tone. Nothing is uploaded, so you can do this with unreleased material without thinking twice about it.
Use it with the other free checks
The loudness meter tells you how the file was finished. The spectrogram viewer shows how the frequency content is distributed and where a codec cut it. The metadata checker reads the tags for any written claim about origin. And the batch checker runs the detector across a whole folder when the question is about a catalogue rather than a single track.
For the theory behind these numbers — how K-weighting and gating work, what streaming normalisation actually does to your master, and why "over-compressed" is a weak argument about origin — read LUFS, loudness and AI music.
Questions about loudness measurement
- What LUFS should I master to for streaming?
- Aim for a master that sounds right rather than a number, then check where it lands. Most platforms normalise playback toward roughly -14 LUFS, and Apple Music toward about -16, so anything between about -14 and -9 LUFS behaves predictably. Mastering far louder than that only means the platform turns it down again, minus the dynamics you gave up.
- Does loudness tell you whether a song is AI-generated?
- No. Generated tracks usually arrive loudness-maximised because they were trained on commercial masters, but so is most commercial music. Crushed dynamics are evidence about mastering, not origin, and treating them as an AI tell is the single most common way people get false positives.
- Is this a true-peak meter?
- It reports sample peak, not oversampled true peak. Inter-sample peaks after lossy encoding can read a few tenths of a decibel higher, which is why delivery guidance usually asks for about 1 dB of headroom rather than exactly 0 dBFS.
- How is the measurement calculated?
- It follows the ITU-R BS.1770 method: two-stage K-weighting, 400 ms blocks with 75 percent overlap, an absolute gate at -70 LUFS and a relative gate 10 LU below the ungated mean. Loudness range comes from gated 3-second short-term values, using the 10th and 95th percentiles as EBU R 128 specifies.
- Is my audio uploaded?
- No. The file is decoded and measured by your own browser and discarded when you close the tab. Nothing is transmitted, stored or logged, which makes it safe for unreleased masters.
- Why does my reading differ slightly from my DAW meter?
- Small differences come from decoding, from files longer than five minutes being measured on a bounded excerpt, and from sample peak versus true peak. Expect agreement within a few tenths of a LU on the integrated figure; if the gap is larger, check that both tools are measuring the same file at the same bit depth.